Rapid prediction of reducing sugar concentration during microcrystalline cellulose hydrolysis using near-infrared spectroscopy and spectral self-supervised network modeling.

Abstract

Monitoring reducing sugars during microcrystalline cellulose (MCC) hydrolysis is important for controlling enzymatic conversion and optimizing cellulose-derived food ingredient production. Near-infrared (NIR) spectroscopy provides a rapid and non-destructive analytical tool, but conventional calibration models usually require large labeled datasets, limiting their practical application under small-sample conditions. In this study, we developed a spectral self-supervised network (SSSN) for label-efficient prediction of reducing sugar concentration during MCC hydrolysis. The framework uses unlabeled NIR spectra for self-supervised pretraining and a limited number of labeled samples for supervised fine-tuning. On the independent test set, SSSN achieved an R2 of 0.9488 and an RMSEP of 0.0867 mg/mL, outperforming partial least squares and support vector machine models. Ablation analysis further showed that the attention module improved both prediction performance and model interpretability by focusing on informative spectral regions associated with reducing sugar variation. These results demonstrate that self-supervised NIR spectroscopy can provide a rapid, non-destructive, and data-efficient strategy for monitoring MCC hydrolysis and supporting quality control in food-related cellulose processing.

Abstract as supplied by the indexing source. © the authors / publisher.